WO2023245827A1 - Method for identifying tissue sources of mesenchymal stem cells in sample and use thereof - Google Patents

Abstract

Provided are a method for constructing a model for identifying tissue sources of mesenchymal stem cells (MSCs), a method and device for identifying tissue sources of MSCs in a sample, and the use of a reagent, which is useful for determining a biomarker level in a sample, in the preparation of a kit.

Description

A method for identifying the tissue origin of mesenchymal stem cells in a sample and its use Technical field

The present application relates to the fields of biology and medicine, specifically to a method of constructing a model for identifying the tissue origin of mesenchymal stem cells (MSCs), a method and device for identifying the tissue origin of MSCs in a sample, and a method for determining the tissue origin of MSCs. Use of reagents for biomarker levels in samples in preparing kits.

Background technique

Human mesenchymal stem cells (hMSCs) are a type of pluripotent adult stem cells with the potential to differentiate into mesodermal lineage cells, and have strong immune regulation, anti-apoptosis, anti-fibrosis, and promotion of Tissue repair and regeneration. Because hMSCs exist in a variety of tissues in the body and are easy to isolate and culture in vitro, hMSCs have high clinical application value. There are currently more than 100 clinical studies in China aimed at studying the safety and effectiveness of MSCs, involving indications such as osteoarthritis, graft-versus-host disease (GvHD), diabetes, premature ovarian failure, etc.

hMSCs were initially isolated and identified in bone marrow. Subsequent studies have shown that hMSCs are widely present in various tissues of the human body, such as adipose tissue, dental tissues such as dental pulp and dental follicles, hair follicles, and perinatal tissues such as fetal umbilical cord and placenta. Many studies have observed that hMSCs derived from different tissues have great differences in cell properties in addition to their different origins. For example, hMSCs derived from bone marrow have strong osteogenic differentiation ability and weak proliferation ability, while hMSCs derived from fat have strong adipogenic differentiation ability, proliferation ability, and stronger IDO1 activity; hMSCs derived from perinatal period have the strongest The proliferation ability, osteogenic and adipogenic differentiation abilities are weak (Front. Med., 20 September 2021 | https://doi.org/10.3389/fmed.2021.728496). In addition to the reported property differences, omics studies have shown that hMSCs from different tissue sources have unique transcriptome expression profiles (Biotechnol Lett.2020 Jul;42(7):1287-1304.doi:10.1007/s10529-020- 02898-x.Epub 2020 May 5.). With the gradual deepening of people's understanding of the biological properties of hMSCs derived from different tissues, and the accumulation of clinical research data for various clinical indications, it is more reasonable and more reasonable to purposefully select MSCs derived from appropriate tissues for the treatment of corresponding diseases. efficient.

It is of great significance to establish a tissue source-specific identification method for hMSCs. First, the tissue origin of hMSCs cell preparations or intermediate cell banks for clinical use can currently only be traced through collection and preparation records. There is no effective method for identification during the quality control process. Once confusion or cross-contamination occurs, it will not be correctly identified. . Second, regulatory agencies/laboratories responsible for quality review of stem cell products also need to use test data to identify and review the tissue source of hMSCs submitted by the production unit for inspection. Third, some researchers are trying to induce differentiation of pluripotent stem cells into cell products with similar or identical properties to hMSCs derived from various tissues for treatment research of specific indications. In this case, it becomes difficult to identify the tissue source of hMSCs. Particularly important.

However, there is no internationally available method for the tissue source-specific identification of hMSCs. Although the International Society for Cell Therapy (ISCT) proposed minimum standards for hMSCs in 2006, this definition is non-specific and fails to address the differences between hMSCs from different tissues and between hMSCs and fibroblasts. Some subsequent studies have tried to study the characteristic identification methods of hMSCs derived from different tissues through surface marker molecules, transcriptomic expression profiles, secretome characteristics, etc. However, no effective method has been established yet. On the one hand, many marker molecules themselves are not It has very good specificity. For example, CD29 is usually considered a surface marker of adipose stem cells, but it is also highly expressed in placenta-derived hMSCs. CD146 is expressed in both bone marrow MSCs and umbilical cord hMSCs. CD271 is expressed in bone marrow and adipose-derived hMSCs. (Stem Cells, Volume 32, Issue 6, June 2014, Pages 1408–1419,). Secondly, some individual studies have tried to explore the characteristic protein expression profile or secretion profile of tissue sources, but the research is often limited by the materials used. It is limited to comparing hMSCs derived from one or two tissues and cannot cover the various hMSCs commonly used in clinical research.

Therefore, there is a need to provide a method that can accurately identify the tissue origin of hMSCs from different tissue sources commonly used in clinical research.

Contents of the invention

This application conducted transcriptome sequencing of hMSCs and used machine learning methods to screen out a combination of biomarkers that can identify the tissue source of hMSCs. The machine learning method was used to train and classify the expression levels of biomarker genes of 137 hMSCs and their tissue sources. Verification, thereby constructing a machine learning model for identifying the tissue source of hMSCs based on a combination of biomarkers. This model can accurately identify the tissue origin of hMSCs from different tissues commonly used in clinical research.

Therefore, in a first aspect, the present application provides a method of constructing a model for identifying the tissue origin of mesenchymal stem cells (MSCs), which includes:

Step (1): Provide n strains of MSCs derived from different tissues, and collect transcriptome sequencing information of the MSCs, where n is an integer greater than or equal to 10;

Step (2): Obtain the mRNA information from the transcriptome sequencing information;

Step (3): Obtain genes with TPMmax greater than 10 from the mRNA information;

Step (4): Use the expression level of the gene obtained in step (3) as a feature vector, filter the feature vector through a machine learning method, and obtain the target feature vector;

Step (5): Use the expression amount of the target feature vector to train the machine learning model to build a model for identifying the tissue source of mesenchymal stem cells (MSCs).

In certain embodiments, in step (4), the expression level of the gene is the TPM value of the gene.

In some embodiments, in step (5), the expression amount of the target feature vector is the TPM value of the target feature vector.

In some embodiments, in step (5), 55% to 95% of samples are randomly extracted from the n strains of MSCs derived from different tissues as a training set, and the target feature vector of the training set is used to train the machine learning model Perform training to build a model for identifying the tissue source of mesenchymal stem cells (MSCs).

In some embodiments, the method further includes step (6): using the MSCs extracted outside the training set as a test set, and testing the machine learning model using the target feature vector of the test set to determine the accuracy of the model. degree, sensitivity and specificity.

In certain embodiments, the machine learning model is selected from Lasso regression, ridge regression, support vector machine, or linear discriminant.

In certain embodiments, the machine learning model is Lasso regression.

In some embodiments, in step (5), 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90 are randomly extracted from the n strains of MSCs derived from different tissues. % or 95% of the samples serve as the training set.

In certain embodiments, the target feature vector includes the following genes or includes transcripts of the following genes: ACVRL1, ARMC9, BCHE, CD55, EBP, FN1, FST, HOTAIRM1, LIMK2, MECOM, METTL26, MSX1, NBPF3 , NECTIN3, NRXN2, PDE5A, RIN3, RPA2, RSL24D1, TSSC2 and ZIC1.

In certain embodiments, the target feature vector is selected from the following genes or a transcript selected from the following genes: ACVRL1, ARMC9, BCHE, CD55, EBP, FN1, FST, HOTAIRM1, LIMK2, MECOM, METTL26, MSX1 , NBPF3, NECTIN3, NRXN2, PDE5A, RIN3, RPA2, RSL24D1, TSSC2, ZIC1, or any combination thereof.

In certain embodiments, n is an integer between 10 and 50, an integer between 51 and 100, an integer between 101 and 150, an integer between 151 and 200, and an integer between 201 and 250. , an integer between 251 and 300, an integer between 301 and 500, or an integer between 501 and 1000.

In certain embodiments, the source of the n strains of MSCs derived from different tissues is selected from bone marrow, umbilical cord, placenta or parts thereof (e.g., placental amniotic membrane), fat, dental pulp, hair follicles, skin, blood, or any of them combination.

In certain embodiments, the MSCs are MSCs derived from a mammal (eg, mouse, human).

In certain embodiments, the MSCs are human-derived MSCs (hMSCs).

In certain embodiments, the gene ACVRL1 has an Entrez Gene ID of 94.

In certain embodiments, the Entrez Gene ID of ARMC9 is 80210.

In certain embodiments, the gene BCHE has an Entrez Gene ID of 590.

In certain embodiments, the gene CD55 has an Entrez Gene ID of 1604.

In certain embodiments, the Entrez Gene ID of the gene EBP is 10682.

In certain embodiments, the gene FN1 has an Entrez Gene ID of 2335.

In certain embodiments, the Entrez Gene ID of the gene FST is 10468.

In certain embodiments, the Entrez Gene ID of the gene HOTAIRM1 is 100506311.

In certain embodiments, the gene LIMK2 has an Entrez Gene ID of 3985.

In certain embodiments, the Entrez Gene ID of MECOM is 2122.

In certain embodiments, the Entrez Gene ID of the gene METTL26 is 84326.

In certain embodiments, the gene MSX1 has an Entrez Gene ID of 4487.

In certain embodiments, the Entrez Gene ID of the gene NBPF3 is 84224.

In certain embodiments, the Entrez Gene ID of NECTIN3 is 25945.

In certain embodiments, the Entrez Gene ID of the gene NRXN2 is 9379.

In certain embodiments, the Entrez Gene ID of the gene PDE5A is 8654.

In certain embodiments, the Entrez Gene ID of the gene RIN3 is 79890.

In certain embodiments, the gene RPA2 has an Entrez Gene ID of 6118.

In certain embodiments, the Entrez Gene ID of the gene RSL24D1 is 51187.

In certain embodiments, the Entrez Gene ID of TSSC2 is 650368.

In certain embodiments, the gene ZIC1 has an Entrez Gene ID of 7545.

In certain embodiments, the Ensembl Gene ID of the gene ACVRL1 is ENSG00000139567.

In certain embodiments, the Ensembl Gene ID of the gene ARMC9 is ENSG00000135931.

In certain embodiments, the Ensembl Gene ID of the gene BCHE is ENSG00000114200.

In certain embodiments, the Ensembl Gene ID of the gene CD55 is ENSG00000196352.

In certain embodiments, the Ensembl Gene ID of the gene EBP is ENSG00000147155.

In certain embodiments, the Ensembl Gene ID of gene FN1 is ENSG00000115414.

In certain embodiments, the Ensembl Gene ID of the gene FST is ENSG00000134363.

In certain embodiments, the Ensembl Gene ID of the gene HOTAIRM1 is ENSG00000233429.

In certain embodiments, the Ensembl Gene ID of the gene LIMK2 is ENSG00000182541.

In certain embodiments, the Ensembl Gene ID of the gene MECOM is ENSG00000085276.

In certain embodiments, the Ensembl Gene ID of the gene METTL26 is ENSG00000130731.

In certain embodiments, the Ensembl Gene ID of the gene MSX1 is ENSG00000163132.

In certain embodiments, the Ensembl Gene ID of the gene NBPF3 is ENSG00000142794.

In certain embodiments, the Ensembl Gene ID of the gene NECTIN3 is ENSG00000177707.

In certain embodiments, the Ensembl Gene ID of the gene NRXN2 is ENSG00000110076.

In certain embodiments, the Ensembl Gene ID of the gene PDE5A is ENSG00000138735.

In certain embodiments, the Ensembl Gene ID of the gene RIN3 is ENSG00000100599.

In certain embodiments, the Ensembl Gene ID of the gene RPA2 is ENSG00000117748.

In certain embodiments, the Ensembl Gene ID of the gene RSL24D1 is ENSG00000137876.

In certain embodiments, the Ensembl Gene ID of the gene TSSC2 is ENSG00000223756.

In certain embodiments, the Ensembl Gene ID of the gene ZIC1 is ENSG00000152977.

On the other hand, the present application provides a machine learning model, which is constructed by the method as described above.

In certain embodiments, the machine learning model is used to identify the tissue of origin of one or more MSCs in a sample (e.g., bone marrow, umbilical cord, placenta or portion thereof (e.g., placental amniotic membrane), fat, dental pulp, hair follicles , skin, blood, or any combination thereof).

In another aspect, the present application provides the use of a machine learning model as previously described to identify the tissue origin of one or more MSCs in a sample.

On the other hand, the present application provides a method for identifying the tissue origin of MSCs in a sample, including:

Step (a): Provide the expression level of the target feature vector of MSCs in the sample, and the target feature vector includes the following genes or the transcript products of the following genes: ACVRL1, ARMC9, BCHE, CD55, EBP, FN1, FST, HOTAIRM1, LIMK2, MECOM, METTL26, MSX1, NBPF3, NECTIN3, NRXN2, PDE5A, RIN3, RPA2, RSL24D1, TSSC2 and ZIC1;

Step (b): Input the expression of the target feature vector into the machine learning model constructed as before to identify the tissue source of the MSCs in the sample.

In certain embodiments, in step (a), the expression level is a TPM value.

In certain embodiments, the TPM value is obtained by transcriptome sequencing.

In certain embodiments, the target feature vector includes the following genes or includes proteins expressed by the following genes: ACVRL1, ARMC9, BCHE, CD55, EBP, FN1, FST, HOTAIRM1, LIMK2, MECOM, METTL26, MSX1 , NBPF3, NECTIN3, NRXN2, PDE5A, RIN3, RPA2, RSL24D1, TSSC2 and ZIC1.

In certain embodiments, the sample contains one or more MSCs.

In some embodiments, in the above step (a), transcriptome sequencing is performed on the MSCs in the sample to obtain the expression level of the target feature vector of the MSCs in the sample; or, in the above step (a) ), the expression level of the target feature vector of MSCs in the sample is obtained by performing expression profile chip detection, single cell transcriptome sequencing, RT-qPCR measurement, and digital PCR measurement on the target feature vector of MSCs in the sample.

In certain embodiments, the tissue source of the MSCs is selected from bone marrow, umbilical cord, placenta or parts thereof (e.g., placental amniotic membrane), fat, dental pulp, hair follicles, tissue from other parts of the placenta, skin, blood, or its random combination.

In certain embodiments, the MSCs are MSCs derived from a mammal (eg, mouse, human).

In certain embodiments, the MSCs are human-derived MSCs (hMSCs).

In certain embodiments, the sample contains a proportion of adipose hMSCs of greater than or equal to 30%.

In certain embodiments, the sample contains bone marrow hMSCs in a proportion of greater than or equal to 40%.

In certain embodiments, the sample contains dental pulp hMSCs in a proportion of greater than or equal to 40%.

In certain embodiments, the sample contains hair follicle hMSCs in a proportion of greater than or equal to 30%.

In certain embodiments, the sample contains umbilical cord hMSCs in a proportion of greater than or equal to 20%.

In certain embodiments, the sample contains placenta-amniotic hMSCs in a proportion of greater than or equal to 40%.

In another aspect, the present application provides a device for identifying the tissue source of mesenchymal stem cells, including:

memory configured to store instructions;

A processor is coupled to the memory, and the processor is configured to execute the method as described above based on instructions stored in the memory.

On the other hand, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and when the instructions are executed by a processor, the method as described above is implemented.

In another aspect, the present application provides a kit for identifying the tissue origin of one or more MSCs in a sample, the kit comprising a reagent for determining the level of a biomarker in the sample, the biomarker The compounds include ACVRL1, ARMC9, BCHE, CD55, EBP, FN1, FST, HOTAIRM1, LIMK2, MECOM, METTL26, MSX1, NBPF3, NECTIN3, NRXN2, PDE5A, RIN3, RPA2, RSL24D1, TSSC2 and ZIC1.

In certain embodiments, the level of the biomarker is the protein or mRNA level of the biomarker.

In certain embodiments, the MSCs are MSCs derived from a mammal (eg, mouse, human).

In certain embodiments, the MSCs are human-derived MSCs (hMSCs).

In another aspect, the present application provides the use of a reagent for determining the level of a biomarker in a sample in the preparation of a kit for identifying the tissue origin of one or more MSCs in the sample; wherein, the The above biomarkers include ACVRL1, ARMC9, BCHE, CD55, EBP, FN1, FST, HOTAIRM1, LIMK2, MECOM, METTL26, MSX1, NBPF3, NECTIN3, NRXN2, PDE5A, RIN3, RPA2, RSL24D1, TSSC2 and ZIC1.

In certain embodiments, the level of the biomarker is the protein or mRNA level of the biomarker.

In certain embodiments, the MSCs are MSCs derived from a mammal (eg, mouse, human).

In certain embodiments, the MSCs are human-derived MSCs (hMSCs).

Definition of Terms

In this disclosure, unless otherwise stated, scientific and technical terms used herein have the meanings commonly understood by those skilled in the art. Moreover, the cell culture, molecular genetics, nucleic acid chemistry, and immunology laboratory procedures used in this article are routine procedures widely used in the corresponding fields. Meanwhile, in order to better understand the present disclosure, definitions and explanations of relevant terms are provided below.

As used herein, the term "sample" refers to a biological sample obtained from a subject, which sample may be a sample that contains or is presumed to contain human mesenchymal stem cells.

As used herein, the terms "machine learning model" or "machine learning method" or "statistical learning method" have the same meaning and may be used interchangeably. It refers to a set of parameters and functions that can establish a corresponding training model through the measured features (target feature vectors) in the training sample. In certain embodiments, the training model can learn from training samples during training with optimized parameters to provide the best quality measure (eg, accuracy) for classifying new samples. In certain embodiments, the parameters and functions may be a collection of linear algebraic operations, nonlinear algebraic operations, and tensor algebraic operations. In certain embodiments, the parameters and functions may include statistical functions, tests, and probability models. In certain embodiments, the measured characteristic in the training sample is the expression of a gene.

As used herein, the term "specificity" refers to the proportion of actual negatives that are themselves correctly identified.

As used herein, the term "sensitivity" refers to the proportion of actual positives that are themselves correctly identified.

As used herein, the term "transcriptome sequencing" or "RNA-seq" refers to the rapid and comprehensive acquisition of almost all specific cells or tissues of a species in a certain state through a sequencing platform (e.g., a second-generation sequencing platform). All transcripts and gene sequences. It can be used to study gene expression, gene function, structure, alternative splicing, and prediction of new transcripts. Usually, in the analysis of transcriptome sequencing, there are three classic values, namely count, FPKM and TPM values.

As used in this article, the term "count" refers to the total number of reads in the sequencing data that are mapped to a certain gene, that is, the measured reads are mapped to the reference genome, and then The software calculates the total number of reads that map to the gene.

As used in this article, the term "FPKM (fragments per kilobase million)" refers to the number of aligned fragments of a gene, normalized to sequencing depth, and then normalized to gene length. , to eliminate the impact of sequencing depth and gene length on the results between different sequencing samples.

As used in this article, the term "TPM (transcripts per million)" refers to the number of aligned fragments of a certain gene. The gene length is first normalized, and then the sequencing depth is normalized. , to eliminate the impact of sequencing depth and gene length on the results between different sequencing samples. In some embodiments, TPM can be used as a measure of gene expression.

As used herein, the term "TPMmax" refers to the maximum TPM value of a gene in a set of samples.

beneficial effects

This application provides a model for identifying the tissue origin of mesenchymal stem cells (MSCs) and a method for constructing the model, which can accurately identify the tissue origin of MSCs from different tissues commonly used in clinical research. The model has been verified multiple times with training sets, test sets and external data sets, and the accuracy, sensitivity and specificity can all reach 95% and above (even as high as 100%). In addition, the model established in this application can also identify the tissue sources of various mixed mesenchymal stem cells in the sample, and the accuracy, sensitivity and specificity can also reach 100%, which has high clinical application value.

Description of the drawings

Figure 1 shows the detection results of the accuracy, sensitivity and specificity of the machine learning model on the training set in Example 2.

Figure 2 shows the detection results of the accuracy, sensitivity and specificity of the machine learning model on the test set in Example 2.

Figure 3 shows the detection results of the accuracy, sensitivity and specificity of the machine learning model on the external data set in Example 2.

Figure 4 shows the prediction ability of the machine learning model for mixed cells in Example 3. Figure 4A shows the detection results of two different sources of hMSCs in simulated mixed sample 1, and Figure 4B shows the detection results of two different types of hMSCs in simulated mixed sample 2. The detection results of hMSCs from two different sources. Figure 4C shows the detection results of hMSCs from two different sources in simulated mixed sample 3.

Figure 5 shows the detection results of hMSCs from three different sources in simulated mixed sample 4 by the machine learning model in Example 3.

Detailed ways

The invention will now be described with reference to the following examples which are intended to illustrate but not to limit the invention. Unless otherwise indicated, the experiments and methods described in the examples were performed essentially according to conventional methods well known in the art and described in various references.

Techniques, methods and equipment known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, such techniques, methods and equipment should be considered part of the authorized specification. In all examples shown and discussed herein, any specific values are to be construed as illustrative only and not as limiting. Accordingly, other examples of the exemplary embodiments may have different values.

In addition, if the specific conditions are not specified in the examples, the conventional conditions or the conditions recommended by the manufacturer shall be followed. If the manufacturer of the reagents or instruments used is not indicated, they are all conventional products that can be purchased commercially. Those skilled in the art will appreciate that the examples describe the invention by way of example and are not intended to limit the scope of the invention as claimed. All publications and other references mentioned herein are incorporated by reference in their entirety.

Example 1. Test materials and equipment

1. A total of 137 hMSCs cells were used to establish the machine learning model, all from the Cell Resource Preservation Research Center of the China Institute for Food and Drug Control. The tissue source and quantity are shown in Table 1 below.

Table 1. hMSCs cells used to build machine learning models

2. A total of 99 hMSCs cell data were collected as an external data set for further testing. The tissue sources and quantities are shown in Table 2 below.

Table 2. hMSCs cells from external data sets

3. The specific materials and equipment used in the test are shown in Table 3 below.

Table 3. Materials and equipment

Example 2. Establishment of machine learning model

1. Establish a machine learning model

1. Transcriptome sequencing

RNA was extracted from the 137 hMSCs cells in Example 1 using the Trizol method, and the cDNA library construction kit in Table 3 was used. The extracted RNA was reverse transcribed into cDNA and a cDNA library was established. Transcription was performed using the sequencing kit in Table 3. Group sequencing was performed to obtain transcriptome sequencing information of 137 hMSCs cells.

2. Data analysis after transcriptome sequencing

After transcriptome sequencing, each sample obtained approximately 6G of cleanbase. The analysis process is shown in Table 4 below.

Table 4. Transcriptome sequencing process

After the above analysis, the gene transcription expression levels of 137 hMSCs cells were obtained, including the count value of transcripts, FPKM and TPM.

3. Statistical learning data preprocessing

Preprocessing software: R (ver=4.1.3), R package tidyverse (ver=1.3.1).

The mRNA was filtered out from the transcripts based on Official Symbol recognition, and a total of 38,735 genes were obtained. The R package tidyverse was used to filter out the highly abundant expressed genes (TPMmax>10), and a total of 13,315 genes were obtained.

4. Statistical learning modeling

Statistical learning modeling platform: R; software glmnet and tidymodel.

The 137 hMSCs transcriptome data were divided into a training set (70%) and a test set (30%). 13315 genes were used as feature vectors, and lasso regression (10-fold cross-validation) was used to screen the feature vectors. The method is: cv. glmnet(x,y,type.measure="class",nfolds=10,family="multinomial",alpha=1,type.multinomial="grouped"). The 10-fold cross-validation result shows that when λ = 0.02552, the misclassification rate is 0, and the target feature vector can be reduced to 21.

Through the above feature vector screening, 21 target feature vectors, namely 21 genes, were finally obtained, as shown in Table 5.

Table 5. 21 genes screened

serial number	symbol ID	Entrez Gene ID	Ensembl Gene ID
1	ACVRL1	94	ENSG00000139567
2	ARMC9	80210	ENSG00000135931
3	BCHE	590	ENSG00000114200
4	CD55	1604	ENSG00000196352
5	EBP	10682	ENSG00000147155
6	FN1	2335	ENSG00000115414
7	FST	10468	ENSG00000134363
8	HOTAIRM1	100506311	ENSG00000233429
9	LIMK2	3985	ENSG00000182541
10	MECOM	2122	ENSG00000085276
11	METTL26	84326	ENSG00000130731
12	MSX1	4487	ENSG00000163132
13	NBPF3	84224	ENSG00000142794
14	NECTIN3	25945	ENSG00000177707
15	NRXN2	9379	ENSG00000110076
16	PDE5A	8654	ENSG00000138735

17	RIN3	79890	ENSG00000100599
18	RPA2	6118	ENSG00000117748
19	RSL24D1	51187	ENSG00000137876
20	TSSC2	650368	ENSG00000223756
twenty one	ZIC1	7545	ENSG00000152977

Perform lasso regression again with the above 21 genes, and establish a machine learning model.

2. Initial establishment and evaluation of machine learning models

The expression levels (ie, TPM values) of 21 genes in the training set (70% of the total sample size) were input into the machine learning model established above, and the accuracy, sensitivity, and specificity of the model's prediction performance were tested. The accuracy test results are shown in Table 6 and Figure 1.

Table 6. Accuracy of model detection

The results showed that the established machine learning model achieved 100% prediction accuracy for the tissue origin of hMSCs in the training set. Not only that, the results showed that the sensitivity and specificity of the machine learning model were also 100%.

Next, the expression levels of 21 genes in the test set (30% of the total sample size) were input into the machine learning model established above, and the accuracy, sensitivity, and specificity of the model's prediction performance were tested. The accuracy test results are shown in Table 7 and Figure 2.

Table 7. Accuracy of model detection

The results showed that the established machine learning model achieved 100% prediction accuracy for the tissue origin of hMSCs in the test set. Not only that, the results showed that the sensitivity and specificity of the machine learning model were also 100%.

Further, the external data set described in Example 1 (a total of 99 hMSCs) was subjected to transcriptome sequencing as described above, and the expression levels of the 21 genes obtained were input into the machine learning model established above. The model's prediction performance accuracy, sensitivity, and specificity were tested. The accuracy test results are shown in Table 8 and Figure 3.

Table 8. Accuracy of model detection

The results showed that the established machine learning model achieved 100% prediction accuracy for the tissue origin of hMSCs in external datasets. Not only that, the results showed that the sensitivity and specificity of the machine learning model were also 100%.

Example 3. Predictive ability of machine learning model for mixed cells

In practical applications, hMSCs from several different tissue sources may be mixed in the detection sample, so this embodiment simulates the situation in which hMSCs from several different tissue sources are mixed.

First, 1,000,000 (1M), 2,000,000 (2M), ... 10,000,000 (10M) reads were extracted from hMSCs transcriptome sequencing, and the hMSCs sequencing reads from different tissue sources were mixed in different proportions to generate new mixed samples. The specific mixed samples are as follows:

The first set of simulation data includes 11 samples in which adipose-derived hMSCs and bone marrow-derived hMSCs were mixed in different proportions, as shown in Table 9:

Table 9. First set of simulation data

​	Number of sequencing reads of adipose hMSCs	Bone marrow hMSCs sequencing read number
Mixed sample 1	0M	10M
Mixed sample 2	1M	9M
Mixed sample 3	2M	8M
Mixed sample 4	3M	7M
Mixed sample 5	4M	6M
Mixed sample 6	5M	5M
Mixed sample 7	6M	4M
Mixed sample 8	7M	3M
Mixed sample 9	8M	2M

Mixed sample 10	9M	1M
Mixed sample 11	10M	0M

The second set of simulation data includes 11 samples of dental pulp-derived hMSCs and hair follicle-derived hMSCs mixed in different proportions, as shown in Table 10:

Table 10. Second set of simulation data

​	Number of sequencing reads of dental pulp hMSCs	Number of sequencing reads of hair follicle hMSCs
Mixed sample 1	0M	10M
Mixed sample 2	1M	9M
Mixed sample 3	2M	8M
Mixed sample 4	3M	7M
Mixed sample 5	4M	6M
Mixed sample 6	5M	5M
Mixed sample 7	6M	4M
Mixed sample 8	7M	3M
Mixed sample 9	8M	2M
Mixed sample 10	9M	1M
Mixed sample 11	10M	0M

The third set of simulation data includes 11 samples of umbilical cord-derived hMSCs and placental amnion-derived hMSCs mixed in different proportions, as shown in Table 11:

Table 11. The third set of simulation data

​	Umbilical cord hMSCs sequencing read number	Placental amnion hMSCs sequencing read number
Mixed sample 1	0M	10M
Mixed sample 2	1M	9M
Mixed sample 3	2M	8M
Mixed sample 4	3M	7M
Mixed sample 5	4M	6M
Mixed sample 6	5M	5M

Mixed sample 7	6M	4M
Mixed sample 8	7M	3M
Mixed sample 9	8M	2M
Mixed sample 10	9M	1M
Mixed sample 11	10M	0M

The above three groups of mixed samples were subjected to tissue source identification analysis according to the method described above, and the results are shown in Figure 4. The results showed that the different tissue sources of hMSCs in multiple groups of mixed samples were accurately predicted.

Further, hMSCs from three different tissue sources were mixed in different proportions. The specific mixed samples are as follows: The mixed samples contain hMSCs derived from fat, bone marrow and hair follicles. After mixing, 11 mixed samples of the fourth group are obtained, as shown in Table 12:

Table 12. Mixed samples of hMSCs from three different sources

Tissue source identification detection was performed as described above. Figure 5 shows the detection results of hMSCs from three different sources in the fourth group of simulated mixed samples. The results showed that the model established in this application accurately predicted the different tissue origins of multiple hMSCs in mixed samples. Therefore, the model established in this application can be used for the detection of mixed samples of hMSCs (containing one or more hMSCs from different sources).

Example 4. Comparison of different machine learning methods

In this example, in order to compare the impact of different machine learning models on the accuracy of the established model for identifying the tissue source of mesenchymal stem cells (MSCs), 5 different machine learning models/methods were selected, according to the method described in Example 2 Establish the above-mentioned model for identifying the tissue source of MSCs (the only difference between the method used in this embodiment and Example 2 is the use of different machine learning models/methods), and verify the accuracy of the established model for identifying the tissue source of MSCs difference.

The experimental results are shown in Table 9. Compared with the lasso regression method in Example 2 (the identification accuracy of the model established by the lasso regression method for the training set, test set and external data set is 100%), ridge regression, Support vector machines and linear discrimination methods can also achieve high accuracy and can be used as alternative methods to Lasso regression modeling to establish the model for identifying the tissue source of mesenchymal stem cells (MSCs) in this application.

Table 9. Machine learning method used in this embodiment

Although the specific embodiments of the present disclosure have been described in detail, those skilled in the art will understand that various modifications and changes can be made to the details based on all teachings that have been published, and these changes are within the protection scope of the present disclosure. . The entire disclosure is given by the appended claims and any equivalents thereof.

Claims (11)

1. A method of constructing a model for identifying the tissue origin of mesenchymal stem cells (MSCs), which includes:

   Step (1): Provide n strains of MSCs derived from different tissues, and collect transcriptome sequencing information of the MSCs, where n is an integer greater than or equal to 10;

   Step (2): Obtain the mRNA information from the transcriptome sequencing information;

   Step (3): Obtain genes with TPMmax greater than 10 from the mRNA information;

   Step (4): Use the expression level of the gene obtained in step (3) as a feature vector, filter the feature vector through a machine learning method, and obtain the target feature vector;

   Step (5): Use the expression amount of the target feature vector to train the machine learning model to build a model for identifying the tissue source of mesenchymal stem cells (MSCs);

   Preferably, in step (5), 55% to 95% of samples are randomly extracted from the n strains of MSCs derived from different tissues as a training set, and the target feature vector of the training set is used to train the machine learning model to Construct a model to identify the tissue source of mesenchymal stem cells (MSCs);

   More preferably, the method further includes step (6): using the MSCs extracted outside the training set as a test set, and testing the machine learning model using the target feature vector of the test set to determine the accuracy and sensitivity of the model. and specificity;

   Preferably, in step (4), the expression level of the gene is the TPM value of the gene;

   Preferably, in step (5), the expression amount of the target feature vector is the TPM value of the target feature vector.
2. The method of claim 1, wherein the machine learning model is selected from Lasso regression, ridge regression, support vector machine or linear discriminant;

   Preferably, the machine learning model is Lasso regression.
3. The method of claim 1 or 2, wherein the method has one or more characteristics selected from the following:

   (1) In step (5), 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90% or 95% are randomly extracted from the n strains of MSCs derived from different tissues. samples as the training set;

   (2) The target feature vector contains the following genes or the transcript products of the following genes: ACVRL1, ARMC9, BCHE, CD55, EBP, FN1, FST, HOTAIRM1, LIMK2, MECOM, METTL26, MSX1, NBPF3, NECTIN3, NRXN2 , PDE5A, RIN3, RPA2, RSL24D1, TSSC2 and ZIC1;

   (3) The target feature vector is selected from the following genes or the transcript products of the following genes: ACVRL1, ARMC9, BCHE, CD55, EBP, FN1, FST, HOTAIRM1, LIMK2, MECOM, METTL26, MSX1, NBPF3, NECTIN3 , NRXN2, PDE5A, RIN3, RPA2, RSL24D1, TSSC2, ZIC1, or any combination thereof;

   (4) The n is an integer between 10 and 50, an integer between 51 and 100, an integer between 101 and 150, an integer between 151 and 200, an integer between 201 and 250, and 251 and 300. an integer between 301 and 500, or an integer between 501 and 1000;

   (5) The sources of the n strains of MSCs derived from different tissues are selected from bone marrow, umbilical cord, placenta or parts thereof (for example, placental amniotic membrane), fat, dental pulp, hair follicles, skin, blood, or any combination thereof;

   (6) The MSCs are MSCs derived from mammals (for example, mice, humans);

   (7) The MSCs are human-derived MSCs (hMSCs);

   Preferably, the transcription product is selected from rRNA, tRNA, mRNA, or non-coding RNA;

   Preferably, the transcript is mRNA.
4. A machine learning model, the machine learning model is constructed by the method described in any one of claims 1-3;

   Preferably, the machine learning model is used to identify the tissue of origin of one or more MSCs in the sample (e.g., bone marrow, umbilical cord, placenta or part thereof (e.g., placental amniotic membrane), fat, dental pulp, hair follicles, skin, blood , or any combination thereof).
5. Use of the machine learning model of claim 4 in identifying the tissue origin of one or more MSCs in a sample.
6. A method for identifying the tissue origin of MSCs in a sample, including:

   Step (a): Provide the expression level of the target feature vector of MSCs in the sample, and the target feature vector includes the following genes or the transcript products of the following genes: ACVRL1, ARMC9, BCHE, CD55, EBP, FN1, FST, HOTAIRM1, LIMK2, MECOM, METTL26, MSX1, NBPF3, NECTIN3, NRXN2, PDE5A, RIN3, RPA2, RSL24D1, TSSC2 and ZIC1;

   Step (b): Input the expression amount of the target feature vector into the machine learning model constructed in claim 4 to identify the tissue source of the MSCs in the sample;

   Preferably, in step (a), the expression level is the TPM value;

   Preferably, the TPM value is obtained by transcriptome sequencing.
7. The method of claim 6, wherein the sample contains one or more MSCs;

   Preferably, in the above step (a), by performing transcriptome sequencing on the MSCs in the sample, the expression level of the target feature vector of the MSCs in the sample is obtained; or, in the above step (a), by Perform expression profiling chip detection, single cell transcriptome sequencing, RT-qPCR measurement, and digital PCR measurement on the target feature vectors of MSCs in the sample to obtain the expression level of the target feature vector of MSCs in the sample;

   Preferably, the tissue source of the MSCs is selected from bone marrow, umbilical cord, placenta or parts thereof (for example, placental amniotic membrane), fat, dental pulp, hair follicles, tissues from other parts of the placenta, skin, blood, or any combination thereof;

   Preferably, the MSCs are MSCs derived from mammals (e.g., mice, humans);

   Preferably, the MSCs are human-derived MSCs (hMSCs).
8. A device for identifying the tissue origin of mesenchymal stem cells, comprising:

   memory configured to store instructions;

   A processor, coupled to the memory, configured to execute the method according to claim 6 or 7 based on instructions stored in the memory.
9. A computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and when the instructions are executed by a processor, the method according to claim 6 or 7 is implemented.
10. A kit for identifying the tissue origin of one or more MSCs in a sample, the kit comprising reagents for determining the levels of biomarkers in the sample, the biomarkers comprising ACVRL1, ARMC9, BCHE, CD55 , EBP, FN1, FST, HOTAIRM1, LIMK2, MECOM, METTL26, MSX1, NBPF3, NECTIN3, NRXN2, PDE5A, RIN3, RPA2, RSL24D1, TSSC2 and ZIC1;

    Preferably, the level of said biomarker is the protein or mRNA level of said biomarker;

    Preferably, the MSCs are MSCs derived from mammals (e.g., mice, humans);

    Preferably, the MSCs are human-derived MSCs (hMSCs).
11. The use of reagents for determining the level of biomarkers in a sample in the preparation of a kit for identifying the tissue origin of one or more MSCs in the sample; wherein the biomarkers include ACVRL1, ARMC9, BCHE, CD55, EBP, FN1, FST, HOTAIRM1, LIMK2, MECOM, METTL26, MSX1, NBPF3, NECTIN3, NRXN2, PDE5A, RIN3, RPA2, RSL24D1, TSSC2 and ZIC1;

    Preferably, the level of said biomarker is the protein or mRNA level of said biomarker;

    Preferably, the MSCs are MSCs derived from mammals (e.g., mice, humans);

    Preferably, the MSCs are human-derived MSCs (hMSCs).

Images